LLM-Based Test Case Generation in DBMS through Monte Carlo Tree Search

Yujia Chen, Yingli Zhou, Fangyuan Zhang, Cuiyun Gao. 2026. arXiv.org.

What this paper does with SQLancer

SQLancer is cited as the traditional approach whose handcrafted dialect-specific generators and sophisticated oracles work well for a specific system but need substantial manual effort to move to another -- the cost MIST aims to remove. ShQveL, from the same project, informed the choice of LLM baseline. MIST generates SQL test cases for DBMSs with lightweight LLMs guided by Monte Carlo tree search. It addresses two industrial constraints: organisations often must use small models for security and privacy reasons, and those models struggle with proprietary SQL dialects and tend to produce semantically similar queries that plateau in coverage. MIST builds a hierarchical feature tree and uses error feedback to steer generation, improving line coverage by 43.3% over the strongest baseline. Written by claude-opus-5 from the 5 places this paper refers to SQLancer. The quotations below are the paper's own words, stored verbatim when the text was extracted.

How it was classified

uses infrastructure — no

SQLancer is cited, not used; nothing in the mentions describes reusing its code.

extends technique — no

No technique is extended; the citation is background.

compares with — no

No empirical comparison against SQLancer is reported in the mentions.

describes as state of the art — no

The citation does not characterise SQLancer as the state of the art.

SQLancer publications it cites (6)

Bibliography entries that resolved to a SQLancer publication, or to a paper by one of the project's authors. A sentence citing one of these numbers is a reference to SQLancer even when it never writes the name.

#	Entry	Matched as
30	Manuel Rigger and Zhendong Su. 2020. Detecting optimization bugs in database engines via non-optimizing reference engine construction. InESEC/FSE ’20: 28th ACM Joint European Software Engineering Conference and Sympos...	sqlancer publication · NOREC
31	Manuel Rigger and Zhendong Su. 2020. Finding bugs in database systems via query partitioning.Proc. ACM Program. Lang.4, OOPSLA (2020), 211:1–211:30.	sqlancer publication · TLP
32	Manuel Rigger and Zhendong Su. 2020. Testing database engines via pivoted query synthesis. In14th USENIX Symposium on Operating Systems Design and Implementation (OSDI 20). 667–682.	sqlancer publication · PQS
33	Manuel Rigger and Zhendong Su. 2020. Testing Database Engines via Pivoted Query Synthesis. In14th USENIX Symposium on Operating Systems Design and Implementation, OSDI. 667–682.	sqlancer publication · PQS
49	Suyang Zhong and Manuel Rigger. 2025. Scaling Automated Database System Testing.arXiv preprint arXiv:2503.21424(2025).	sqlancer publication
50	Suyang Zhong and Manuel Rigger. 2025. Testing Database Systems with Large Language Model Synthesized Fragments.arXiv preprint arXiv:2505.02012(2025).	sqlancer publication

Every place it refers to SQLancer (5)

5 sentences, each stored verbatim from the extracted text with where it was found and how. “Citation marker” means the sentence names no tool at all and was reached through a reference number that resolved to a SQLancer publication.

Id	Sentence	Found by	Where
M1	Traditional approaches for generating test cases, such as BuzzBee [ 46], SQLsmith [ 34], SQLancer [ 32], can be effective for a specific DBMS; however, there exist many DBMSs designed for different application scenarios, such as DuckDB [ 29] for analytical workloads, PostgreSQL [ 36] for general-purpose OLTP systems...	name motivation	1 Introduction page 1
M2	Adapting these methods to different DBMSs requires substantial manual effort, such as crafting grammar rules and maintaining specific operators, which severely limits their scalability [ 48–50].	citation marker motivation	1 Introduction page 1
M3	2 Studied DBMSs Following prior studies [ 49,50], we evaluate MIST on three widelyused open-source DBMS representing diverse architectural designs and SQL dialect characteristics.	citation marker background	4.2 Studied DBMSs page 6
M4	Inspired by ShQveL [ 50], we adopt the state-of-the-art LLMbased framework Fuzz4All for fair comparison.	name background	4.3 Studied LLMs and Baseline page 7
M5	SQLancer [ 30,31,33] employs handcrafted dialect-specific generators with sophisticated test oracles to detect logic bugs.	name definition	7.1 LLM-based Testing page 11

This page is rendered from _data/papers/paper_arxiv_2603_21530.json, extracted from arxiv. 12 pages, 52 references parsed.